Horama/wow_scraped

Wildlife Scraped

Multi-source wildlife image dataset covering 122 labels (species, sub-species and a few domestic categories).

Each parquet shard is a HuggingFace Image feature so images decode straight to PIL.Image via load_dataset.

Quick stats

• 139,231 images, 122 labels
• last updated: 2026-04-29
• sources: inaturalist, wikimedia, wikipedia, flickr, ddg

Splits (across all sources): train = 97,456 (70.0%) valid = 20,913 (15.0%) test = 20,862 (15.0%)

Source	Images	train	valid	test
inaturalist	121,851	85,288	18,279	18,284
wikimedia	7,692	5,386	1,176	1,130
wikipedia	0	0	0	0
flickr	0	0	0	0
ddg	9,688	6,782	1,458	1,448
TOTAL	139,231	97,456	20,913	20,862

Per-label image counts (122 labels, 139231 images) – click to expand

label	specie	inaturalist	wikimedia	wikipedia	flickr	ddg	train	valid	test	total
addax	addax	746	54	0	0	0	561	119	120	800
agouti_azara	agouti	2365	50	0	0	0	1690	363	362	2415
alpaga	alpaga	523	74	0	0	0	419	89	89	597
amazone_aourou	amazone	840	38	0	0	0	614	131	133	878
ane_commun	ane	2000	68	0	0	0	1448	310	310	2068
ane_somalie	ane	190	52	0	0	317	390	86	83	559
anoa	anoa	99	46	0	0	340	340	72	73	485
antilope_cervicapre	antilope	1485	70	0	0	0	1088	235	232	1555
autruche	autruche	1000	76	0	0	0	753	162	161	1076
bison_amerique	bison	2018	67	0	0	0	1465	310	310	2085
bison_europe	bison	786	72	0	0	0	599	128	131	858
boeuf_ecosse	boeuf	0	59	0	0	868	649	139	139	927
bongo	bongo	257	53	0	0	344	465	95	94	654
cacatoes_rosalbin	cacatoes	1465	75	0	0	0	1079	229	232	1540
calao_abyssinie	calao	1159	81	0	0	0	868	186	186	1240
calopsitte	calopsitte	1127	81	0	0	0	845	181	182	1208
canard	canard	1000	78	0	0	0	754	162	162	1078
capybara	capybara	2000	79	0	0	0	1455	306	318	2079
cariama_huppe	cariama	1688	76	0	0	0	1236	265	263	1764
casoar	casoar	1843	53	0	0	0	1324	292	280	1896
chat	chat	1000	80	0	0	0	756	162	162	1080
cheval_comtois	cheval	1000	63	0	0	0	744	161	158	1063
cheval_przewalski	cheval	478	87	0	0	0	395	82	88	565
chevre_anglo_nubienne	chevre	30	101	0	0	568	486	101	112	699
chevre_naine	chevre	0	50	0	0	637	481	104	102	687
chien	chien	1000	110	0	0	0	777	167	166	1110
chien_buisson	chien_buisson	238	66	0	0	212	360	77	79	516
chouette_effraie	chouette	2000	43	0	0	0	1430	306	307	2043
coati_nez_blanc	coati	1000	75	0	0	0	765	155	155	1075
cobe_croissant	cobe	1000	56	0	0	0	739	159	158	1056
cobe_lechwe	cobe	627	86	0	0	0	498	109	106	713
cochon_d_inde	cochon	587	77	0	0	0	463	98	103	664
cochon_laineux	cochon	1000	92	0	0	322	991	212	211	1414
colobe_guereza	colobe	2334	63	0	0	0	1678	360	359	2397
colombine_turvert	colombine	1000	33	0	0	0	724	155	154	1033
corbeau	corbeau	1000	73	0	0	0	751	161	161	1073
coyote	coyote	1000	95	0	0	0	767	164	164	1095
daim_europe	daim	2000	71	0	0	0	1450	311	310	2071
dhole	dhole	1134	70	0	0	477	1176	255	250	1681
dromadaire	dromadaire	871	77	0	0	0	664	143	141	948
eland_cap	eland	2404	83	0	0	0	1741	374	372	2487
elephant_afrique	elephant	1000	56	0	0	0	739	159	158	1056
emeu	emeu	1000	86	0	0	0	761	163	162	1086
fourmilier	fourmilier	939	81	0	0	0	715	152	153	1020
gazelle_perse	gazelle	1232	82	0	0	0	920	196	198	1314
gibbon_bonnet	gibbon	329	45	0	0	254	437	93	98	628
girafe_kordofan	girafe	1189	56	0	0	374	1133	244	242	1619
gnou_bleu	gnou	1000	69	0	0	0	746	167	156	1069
gnou_queue_blanche	gnou	1183	71	0	0	0	873	199	182	1254
gorille_plaines	gorille	712	71	0	0	0	549	117	117	783
grand_eclectus	eclectus	384	24	0	0	299	495	106	106	707
grand_hocco	hocco	754	28	0	0	0	548	117	117	782
grand_koudou	koudou	2000	74	0	0	0	1452	312	310	2074
grue_couronnee_grise	grue	1114	30	0	0	0	797	169	178	1144
guepard	guepard	1000	83	0	0	0	762	161	160	1083
hapalemur_lac	hapalemur	918	2	0	0	0	638	132	150	920
hippopotame	hippopotame	2000	92	0	0	0	1464	314	314	2092
hippotrague_noir	hippotrague	1345	90	0	0	0	1007	215	213	1435
hocco_daubenton	hocco	33	1	0	0	239	191	41	41	273
hyene_tachetee	hyene	1000	60	0	0	0	742	159	159	1060
kangourou_roux	kangourou	1000	54	0	0	0	738	158	158	1054
lapin_geant_papillon	lapin	2000	50	0	0	0	1437	308	305	2050
lemur_couronne	lemur	563	12	0	0	0	401	87	87	575
lemur_noir	lemur	875	83	0	0	0	671	144	143	958
lemur_ventre_rouge	lemur	640	22	0	0	0	463	100	99	662
lion_afrique	lion	1000	58	0	0	0	740	161	157	1058
loriquet_arc_en_ciel	loriquet	1000	85	0	0	0	759	162	164	1085
loup_arctique	loup	296	47	0	0	396	516	111	112	739
loup_criniere	loup_criniere	891	84	0	0	0	683	146	146	975
loup_mackenzie	loup	348	30	0	0	285	460	107	96	663
loutre_asie	loutre	1710	73	0	0	0	1247	266	270	1783
lynx_carpates	lynx	2057	64	0	0	0	1486	318	317	2121
macaque_tonkean	macaque_noir	1159	49	0	0	223	999	221	211	1431
mangabey_dore	mangabey	859	86	0	0	207	807	173	172	1152
maki_catta	maki	1358	80	0	0	0	1006	217	215	1438
maki_vari_noir	maki	1119	54	0	0	0	821	176	176	1173
maki_vari_roux	maki	842	31	0	0	0	617	129	127	873
mara	mara	1408	80	0	0	0	1043	223	222	1488
milan_noir	milan	1000	100	0	0	0	769	164	167	1100
moufette	moufette	1000	66	0	0	0	746	161	159	1066
mouton_somalie	mouton	1000	55	0	0	615	1170	250	250	1670
mouton_valachie	mouton	0	12	0	0	609	434	94	93	621
muntjac_chine	muntjac	2000	49	0	0	0	1434	308	307	2049
nandou	nandou	1000	48	0	0	0	736	156	156	1048
oie	oie	1000	73	0	0	0	753	160	160	1073
oryx_algazelle	oryx	1196	74	0	0	0	889	190	191	1270
oryx_arabie	oryx	852	79	0	0	0	651	140	140	931
ouistiti_pygmee	ouistiti	1097	76	0	0	0	818	178	177	1173
ours_baribal	ours	1132	69	0	0	0	840	180	181	1201
ours_lunettes	ours	491	73	0	0	0	393	84	87	564
panda_roux	panda	611	62	0	0	0	473	100	100	673
panthere_chine	panthere	525	8	0	0	246	545	117	117	779
panthere_neige	panthere	471	53	0	0	288	572	122	118	812
patas	patas	1292	70	0	0	0	953	205	204	1362
pelican_blanc	pelican	1000	64	0	0	0	745	160	159	1064
perruche_patagonie	perruche	541	73	0	0	0	427	91	96	614
phacochere	phacochere	1000	63	0	0	0	742	164	157	1063
pigeon_madagascar	pigeon	976	11	0	0	0	691	148	148	987
raton_laveur	raton_laveur	1000	70	0	0	0	749	160	161	1070
renard_polaire	renard	1000	83	0	0	0	758	166	159	1083
rhinoceros_blanc	rhinoceros	1000	81	0	0	0	759	162	160	1081
saimiri_perou	saimiri	2034	68	0	0	0	1474	314	314	2102
saki_face_blanche	saki	1206	62	0	0	0	889	190	189	1268
serval	serval	455	54	0	0	227	518	109	109	736
sitatunga	sitatunga	319	48	0	0	286	457	96	100	653
springbok	springbok	647	71	0	0	0	500	115	103	718
suricate	suricate	478	73	0	0	0	385	82	84	551
tamarin_goeldi	tamarin	249	47	0	0	490	548	121	117	786
tamarin_lion	tamarin	407	5	0	0	276	477	102	109	688
tamarin_pinche	tamarin	1288	92	0	0	0	964	210	206	1380
tamarin_roux	tamarin	655	51	0	0	0	492	101	113	706
tapir_terrestre	tapir	833	70	0	0	0	632	135	136	903
tigre_siberie	tigre	1102	80	0	0	0	826	178	178	1182
titi_roux	titi	513	0	0	0	289	561	121	120	802
urubu_tete_rouge	urubu	1000	81	0	0	0	757	162	162	1081
vanneau_soldat	vanneau	1000	36	0	0	0	725	157	154	1036
vautour_fauve	vautour	1000	86	0	0	0	762	162	162	1086
vautour_ruppell	vautour	1234	10	0	0	0	870	186	188	1244
vigogne	vigogne	1000	94	0	0	0	766	164	164	1094
wallaby_bennett	wallaby	1000	79	0	0	0	755	162	162	1079
watusi	watusi	1000	101	0	0	0	770	170	161	1101
zebre_chapman	zebre	696	76	0	0	0	540	117	115	772
TOTAL		121851	7692	0	0	9688	97456	20913	20862	139231

Porto subset

A curated selection for the Porto zoo project, tracked separately in <src>/porto_metadata.csv. Same schema as the main metadata.csv. The subset contains 42,125 images / 39 labels — most rows overlap with the main metadata, but a few labels are exclusive to porto (not in thoiry_labels.json): aigle_royal, ane_miranda, buffle_eau_asiatique, cerf_philippines, cochon_vietnamien, faisan_dore, girafe, iguane_vert, lapin_domestique, macaque_ours, nyala, pigeon_couronne_occidental, tamarin_empereur, vautour_charognard.

The full label list is available at porto_labels.json.

Per-label image counts (porto subset) – click to expand

label	specie	inaturalist	wikimedia	ddg	train	valid	test	total
agouti_azara	agouti	3093	100	0	2235	478	480	3193
aigle_royal	aigle	10	10	0	16	3	1	20
ane_miranda	ane	2010	78	0	1462	312	314	2088
antilope_cervicapre	antilope	1485	70	0	1088	235	232	1555
autruche	autruche	1000	76	0	752	165	159	1076
buffle_eau_asiatique	buffle	10	9	0	13	3	3	19
capybara	capybara	2000	79	0	1456	312	311	2079
cerf_philippines	cerf	10	2	0	8	3	1	12
cheval_comtois	cheval	1000	63	0	744	160	159	1063
chevre_naine	chevre	0	50	637	481	104	102	687
cobe_lechwe	cobe	627	86	0	499	107	107	713
cochon_d_inde	cochon	587	77	0	465	100	99	664
cochon_vietnamien	cochon	1000	92	322	990	212	212	1414
emeu	emeu	1000	86	0	760	163	163	1086
faisan_dore	faisan	10	10	0	13	5	2	20
girafe	girafe	3189	112	374	2577	549	549	3675
gnou_queue_blanche	gnou	1820	141	0	1373	295	293	1961
guepard	guepard	1000	83	0	758	163	162	1083
hyene_tachetee	hyene	1000	60	0	742	159	159	1060
iguane_vert	iguane	10	9	0	15	3	1	19
lapin_domestique	lapin	2000	50	0	1435	312	303	2050
loutre_asie	loutre	1727	142	0	1307	281	281	1869
macaque_ours	macaque	10	9	0	12	3	4	19
maki_catta	maki	2358	158	0	1759	374	383	2516
maki_vari_roux	maki	846	62	0	636	136	136	908
nyala	nyala	10	10	0	13	4	3	20
panda_roux	panda	611	62	0	472	101	100	673
phacochere	phacochere	1000	63	0	745	160	158	1063
pigeon_couronne_occidental	pigeon	10	10	0	14	4	2	20
pigeon_madagascar	pigeon	992	22	0	710	151	153	1014
rhinoceros_blanc	rhinoceros	1000	81	0	757	163	161	1081
suricate	suricate	478	73	0	385	83	83	551
tamarin_empereur	tamarin	10	10	0	14	4	2	20
tamarin_pinche	tamarin	2288	183	0	1730	371	370	2471
tapir_terrestre	tapir	833	70	0	632	136	135	903
tigre_siberie	tigre	1108	160	0	886	189	193	1268
vautour_charognard	vautour	10	9	0	13	4	2	19
vigogne	vigogne	1000	94	0	766	164	164	1094
wallaby_bennett	wallaby	1000	79	0	756	162	161	1079
TOTAL		38152	2640	1333	29489	6333	6303	42125

Sources

Source	Folder	Code	Default size
iNaturalist	inaturalist/	ina	medium (500 px)
Wikimedia Commons	wikimedia/	wkm	medium (500 px)
Wikipedia	wikipedia/	wkp	medium (500 px)
Flickr	flickr/	fli	medium (500 px)
DuckDuckGo	ddg/	ddg	medium (500 px)

Filtering

• iNaturalist – only Alive annotations (term 17 / value 18); quality grades research, needs_id, casual; wild and captive both kept (recorded in inat_savage). Image sizes restricted to small (240 px) or medium (500 px).
• Flickr / Wikimedia / Wikipedia / DDG – negative-keyword title/URL filter (dead, carcass, skull, bones, taxidermy, trophy, hunt, feces, dung, scat, slaughter, roadkill, …).
• DDG + Wikipedia – inline CLIP alive vs not-alive classification (openai/clip-vit-base-patch32); rejected URLs are persisted in <src>/rejected_urls/<worker>.csv so they are skipped on subsequent runs.
• Relevance filter (post-hoc, scraping.relevance_filter) – species-aware CLIP pass on already-pushed shards: drops rows that fail either an alive threshold or a species-relevance threshold (positive prompts built from the english / scientific / sub-species names). Dropped URLs are appended to <src>/rejected_urls/post_filter.csv so subsequent scrapes skip them. Typical use: cleaning landscapes, prey, or off-species photos from older DDG / Wikipedia shards.

Repository layout

Horama/wow_scraped/
├── inaturalist/
│   ├── 0-000.parquet               # worker 0, shard 0 (~500 MB)
│   ├── 0-001.parquet
│   ├── 3-000.parquet               # worker 3, shard 0
│   ├── …
│   ├── metadata.csv                # consolidated metadata
│   └── metadata/
│       ├── 0-000.csv               # per-shard metadata (raw)
│       └── …
├── wikimedia/  …
├── wikipedia/  …                   # also contains rejected_urls/
├── flickr/     …
└── ddg/        …                   # also contains rejected_urls/

Schema

Each parquet shard:

image:        struct<bytes: binary, path: string>   # HF Image feature
filename:     string    # src3_VV_W_NNNNNN.ext (e.g. ina_00_3_000042.jpg)
label:        string    # canonical label (e.g. loup_arctique)
specie:       string    # broader group (e.g. loup)
sub_specie:   string    # = label (full granularity)
src:          string    # inaturalist | wikimedia | wikipedia | flickr | ddg
licence:      string    # CC0, CC-BY, ..., unknown
author:       string
url:          string    # original source URL
resolution:   string    # "WxH"
split:        string    # train | valid | test (added by scraping.add_split_column)
inat_obs:     string    # iNaturalist observation id (other sources: "")
location:     string    # "lat,lon" when available, else ""
inat_quality: string    # research | needs_id | casual (iNat only)
inat_savage:  string    # "savage" | "captive" (iNat only)

Loading

from datasets import load_dataset

# All shards from one source
ds = load_dataset("Horama/wow_scraped", data_dir="inaturalist", split="train")
print(ds[0]["image"])          # PIL.Image.Image
print(ds[0]["label"], ds[0]["licence"])

# A single shard
ds = load_dataset(
    "Horama/wow_scraped",
    data_files={"train": "inaturalist/0-000.parquet"},
    split="train",
)

Filtering by split

The split assignment (train / valid / test) lives in the per-source metadata.csv only — the parquets themselves don't carry it (so you keep the freedom to re-split without re-uploading the images). Join the metadata to filter:

import csv
from datasets import load_dataset
from huggingface_hub import hf_hub_download

split_lookup: dict[str, str] = {}
csv_path = hf_hub_download("Horama/wow_scraped", repo_type="dataset",
                            filename="inaturalist/metadata.csv")
with open(csv_path) as f:
    for row in csv.DictReader(f):
        split_lookup[row["filename"]] = row["split"]

ds = load_dataset("Horama/wow_scraped", data_dir="inaturalist", split="train")
train = ds.filter(lambda r: split_lookup.get(r["filename"]) == "train")
valid = ds.filter(lambda r: split_lookup.get(r["filename"]) == "valid")
test  = ds.filter(lambda r: split_lookup.get(r["filename"]) == "test")

iNaturalist observations are kept whole (no inat_obs is split between train/valid/test).

Licensing

Per-image license is preserved verbatim in the licence column so you can filter / partition the dataset for either research or production use.

Common values	Allowed for
CC0, Public Domain Mark 1.0, No known copyright restrictions, US Government Work	any (incl. commercial)
CC BY, CC BY-SA	any with attribution
CC BY-NC, CC BY-NC-SA, CC BY-NC-ND, cc-by-nc	research only
unknown (DDG)	research only, attribution undetermined

Reproducing the dataset

The full pipeline (workers, sharding, alive filter, mailer) lives at Horama/WOW_dataset_creation under scraping/.

Citation

If you use this dataset, please credit Horama and the underlying contributors of each photo (see licence and author columns).